LightRAG

About

Lightweight RAG framework with graph indexing. Fast, Pythonic RAG with incremental updates and efficient storage.

• Dual-level graph: local (entity-level) and global (document-level) relationships for multi-granularity retrieval
• Incremental update API that adds new text to the graph without recomputing existing embeddings
• Configurable retrieval modes: 'naive', 'local', 'global', or 'hybrid' to balance speed vs. richness
• Built-in caching for query responses to reduce latency on repeated questions
• Integration adapters for LlamaIndex, LangChain, and direct Pythonic API